System challenges result in imaging advances
George Kotelly Executive Editor
In meeting special application requirements, machine-vision and imaging systems often face performance challenges throughout their range of operation. To capture images, selecting the proper lens and optics usually proves the first major system challenge. To process images, inadequate lighting generally calls for sophisticated hardware and software integration. To analyze complex images, advanced restoration techniques must usually be applied to correct such deficiencies as deblurring.
For many imaging applications, off-the-shelf optics and lenses prove inadequate. Consequently, developers are forced to design custom lenses and optics to meet their vision-system requirements. To help this process, reports editor at large Andy Wilson, several companies are providing modular software packages that can be combined with other computer-aided-design software to analyze, optimize, and visualize optical systems as three-dimensional models. Other software options offer models of standard light sources, libraries of lens examples, and CAD tools (see p. 43).
New imaging applications generally demand novel processes. For example, to inspect the security features incorporated into the newly designed $100, $50, and $20 bills, the US Bureau of Engraving and Printing has installed a custom imaging-inspection system to ensure the quality of its bank-note printing process. Says contributing editor R. Winn Hardin, this CCD-camera and dedicated-processor system is validating the presence of enlarged portraits, embedded fine lines, and security threads (see p. 25).
In many telescope and microscope applications, the generated images are often degraded or out of focus and need image-restoration techniques to highlight specific details. In applications where the cause of blurring is unknown, researchers are investigating blind-image deconvolution techniques in attempts to deblur images. As spotlighted by Andy Wilson, these deconvolution techniques are being enhanced by two development algorithms--maximum-likelihood and non-negativity and support-constrained, recursive inverse filtering (see p. 35).
In using image-segmentation techniques to derive enhanced results, says Peter Eggleston of Amerinex Applied Imaging, segmentation operators might over- or undersegment the imaging data; he describes what can be done to correct this situation through the use of split and merge techniques (see p. 21).